Correlated Input-Dependent Label Noise in Large-Scale Image Classification

TL;DR

This paper introduces a probabilistic model for input-dependent label noise in large-scale image classification, improving accuracy by capturing heteroscedastic uncertainty with a learned covariance structure.

Contribution

It proposes a novel multivariate Normal latent variable approach to model label noise, enhancing classification performance on multiple large datasets.

Findings

Achieved state-of-the-art accuracy on WebVision 1.0.

Significantly improved results on ImageNet and JFT datasets.

Learned covariance structures align with known noise sources.

Abstract

Large scale image classification datasets often contain noisy labels. We take a principled probabilistic approach to modelling input-dependent, also known as heteroscedastic, label noise in these datasets. We place a multivariate Normal distributed latent variable on the final hidden layer of a neural network classifier. The covariance matrix of this latent variable, models the aleatoric uncertainty due to label noise. We demonstrate that the learned covariance structure captures known sources of label noise between semantically similar and co-occurring classes. Compared to standard neural network training and other baselines, we show significantly improved accuracy on Imagenet ILSVRC 2012 79.3% (+2.6%), Imagenet-21k 47.0% (+1.1%) and JFT 64.7% (+1.6%). We set a new state-of-the-art result on WebVision 1.0 with 76.6% top-1 accuracy. These datasets range from over 1M to over 300M training examples and from 1k classes to more than 21k classes. Our method is simple to use, and we provide an implementation that is a drop-in replacement for the final fully-connected layer in a deep classifier.